| Benchmark | DeepSeek-V3 | Qwen3.5 35B A3B |
|---|---|---|
| AA Agentic Index | 1.6 | 44.1 |
| AA Intelligence | 15.4 | 29.9 |
| AA-LCR | 41.3 | 68.3 |
| AA-Omniscience | -37.6 | -48.1 |
| arc_challenge | 95.3 | 95.4 |
| Artificial Analysis Coding Index | 23 | 37 |
| C-Eval | 90.1 | 90.2 |
| critpt | 0 | 0.9 |
| gdpval | 0 | 20.3 |
| GPQA Diamond | 68.4 | 84.5 |
| GSM8K | 96.7 | 90.1 |
| hellaswag | 88.9 | 85.6 |
| HLE | 5.2 | 47.4 |
| HMMT 2025 | 27.5 | 89.2 |
| hmmt_feb_2025 | 29.2 | 89 |
| humaneval | 92.1 | 66.5 |
| IFBench | 41 | 72.5 |
| ifeval | 87.3 | 91.9 |
| LiveCodeBench v6 | 46.9 | 74.6 |
| longbench_v2 | 48.7 | 59 |
| MBPP | 75.4 | 70.8 |
| mmlu | 89.4 | 81.1 |
| mmlu_prox | 70.5 | 81 |
| mmlu_redux | 90.5 | 93.3 |
| MMLU-Pro | 81.2 | 85.3 |
| mmmlu | 79.4 | 85.2 |
| multichallenge | 31.4 | 60 |
| OJBench | 24 | 36 |
| OmniScience Accuracy | 25.4 | 20.1 |
| OmniScience Non-Hallucination | 23.3 | 14.6 |
| OpenBookQA | 95.4 | 44.2 |
| scicode | 35.8 | 37.7 |
| supergpqa | 53.7 | 63.4 |
| SWE-bench Verified | 42 | 70 |
| TauBench V3 - Banking | 4.7 | 4.9 |
| Terminal-Bench 2.1 | 16.9 | 40.8 |
| Terminal-Bench Hard | 15.2 | 26.5 |
| vectara_answer_rate | 97.5 | 99.8 |
| vectara_avg_summary_length | 81.7 | 94.9 |
| vectara_factual_consistency | 93.9 | 89.5 |
| vectara_hallucination_rate ↓ | 6.1 | 10.5 |
| winogrande | 84.9 | 79.2 |
| τ²-Bench Telecom (AA run) | 47.1 | 89.2 |
| τ³-Bench | 4.7 | 68.9 |
Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing.